[pbs-devel] [PATCH proxmox{, -backup} v2 0/7] fix #6665: never mark chunks as bad on s3 client fetch error

[Christian Ebner]

This patch series fixes an issue with chunks being incorrectly marked
as bad in case of s3 client fetch errors. In that case, only increase
the error counter and log the error, but leave the chunk as is.

To increase robusteness, implement a retry logic for all transient s3
client request errors with exponential backoff time. Further,
unconditionally increase the get object request timeout from currently
1 minute to 30 minutes, to greatly reduce possible timeouts on low
download bandwidth connections.

Changes since version 1 (thanks @Wolfgang for offlist feedback):
- Change timeout value to be independent of request retries
- Adapt backoff time to be exponential, increasing initial value to 1 second
- Do not increase tcp keepalive idle time, and rename const to avoid confusion

proxmox:

Christian Ebner (6):
  proxmox-http: add method to share full body as contiguous bytes
  s3-client: drop non-ambiguous mention of chunks in error message
  s3-client: fix unintended match statement being an expression
  s3-client: bump s3 request timeout from 1 minute to 30 minutes
  s3-client: add retry logic for transient client errors
  s3-client: use better fitting name for TCP idle time

 proxmox-http/src/body.rs        |  8 ++++
 proxmox-s3-client/src/client.rs | 74 +++++++++++++++++++++++----------
 2 files changed, 61 insertions(+), 21 deletions(-)


proxmox-backup:

Christian Ebner (1):
  fix #6665: never rename chunks on s3 client fetch errors

 src/backup/verify.rs | 35 +++++++++++++++++++++++------------
 1 file changed, 23 insertions(+), 12 deletions(-)


Summary over all repositories:
  3 files changed, 84 insertions(+), 33 deletions(-)

-- 
Generated by git-murpp 0.8.1